Aiming at the problem of weak fault signals of magnetic coupler bearings and the difficulty of feature extraction, which leads to the low accuracy of fault classification, a rolling bearing fault diagnosis method is proposed, which uses improved beluga whale optimization (IBWO) to optimize variational mode decomposition (VMD), and combines the hybrid model of convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM). Firstly, the IBWO algorithm is used to optimize the two key parameters of VMD (mode number K and penalty factor α). Then, the two optimized parameters are substituted into VMD to obtain K intrinsic mode functions (IMFs). Next, the IMF component with the minimum envelope entropy is selected as the effective IMF component, which is finally input into the CNN-BiLSTM model for fault diagnosis. Experiments are conducted using the public datasets from the Case Western Reserve University and the University of Ottawa, respectively. The results show that the fault identification accuracy of the proposed model can reach more than 95%, which proves that the present diagnosis method has a significant advantage in identification accuracy. The research results can provide a reference for the fault diagnosis of magnetic coupler bearings.